In this jupyter cookbook, we will explore the HAWC+ data cube and describe some of the basic analysis techniques involving imaging polarimetry data.
This cookbook follows the SOFIA press release of 30 Doradus observations: SOFIA Reveals Never-Before-Seen Magnetic Field Details.
The Level 4 reduced data from this program has been released immediately to the public and is available on the SOFIA Data Cycle System (DCS). This notebook will guide the reader through downloading the 30 Doradus data with a walkthrough of basic analysis techniques with python.
HAWC_PLUS from drop-down menuLEVEL_4 from drop-down menuSIMBAD Position buttonSpatial Search Radius to 600 arcsecSubmit buttonGet Selected Data in Current PageRequest Data BundleAfter downloading the SOFIA DCS bundle to your working directory you will want to unzip it, which will produce a directory structure like this:
.
└── sofia_data
├── level4
│ └── p5813
│ └── F0484_HA_POL_7600018_HAWCHWPC_PMP_022-114.fits
└── missions
├── 2018-07-05_HA_F481
│ └── p5827
│ └── F0481_HA_POL_7600012_HAWDHWPD_PMP_050-083.fits
├── 2018-07-07_HA_F483
│ └── p5646
│ └── F0483_HA_POL_7600014_HAWCHWPC_PMP_022-065.fits
├── 2018-07-11_HA_F484
│ └── p5648
│ └── F0484_HA_POL_7600017_HAWCHWPC_PMP_065-114.fits
└── 2018-07-12_HA_F485
└── p5658
├── g1
│ └── F0485_HA_POL_76000110_HAWAHWPA_PMP_043-052.fits
└── g2
└── F0485_HA_POL_7600019_HAWEHWPE_PMP_055-075.fits
Note that each file represents observations with a different filter. However, two observations were made with the same filter (HAWC C, $89\,\mathrm{\mu m}$). These files, F0483_HA_POL_7600014_HAWCHWPC_PMP_022-065.fits and F0484_HA_POL_7600017_HAWCHWPC_PMP_065-114.fits, were combined into one: level4->p5813->F0484_HA_POL_7600018_HAWCHWPC_PMP_022-114.fits.
You can choose to keep the fits files nested, or copy them into one directory.
For the purpose of this basic analysis, though, let us dump all the files into one sofia_data directory:
.
└── sofia_data
├── F0481_HA_POL_7600012_HAWDHWPD_PMP_050-083.fits
├── F0483_HA_POL_7600014_HAWCHWPC_PMP_022-065.fits
├── F0484_HA_POL_7600017_HAWCHWPC_PMP_065-114.fits
├── F0484_HA_POL_7600018_HAWCHWPC_PMP_022-114.fits
├── F0485_HA_POL_76000110_HAWAHWPA_PMP_043-052.fits
└── F0485_HA_POL_7600019_HAWEHWPE_PMP_055-075.fits
For this analysis, we require the standard numpy/scipy/matplotlib stack as well the astropy and aplpy modules.
With just a few lines of code, we can explore the HAWC+ fits data cubes and plot the images.
from astropy.io import fits
efile = 'sofia_data/F0485_HA_POL_7600019_HAWEHWPE_PMP_055-075.fits'
dfile = 'sofia_data/F0481_HA_POL_7600012_HAWDHWPD_PMP_050-083.fits'
cfile = 'sofia_data/F0484_HA_POL_7600018_HAWCHWPC_PMP_022-114.fits'
afile = 'sofia_data/F0485_HA_POL_76000110_HAWAHWPA_PMP_043-052.fits'
hawc = fits.open(afile)
hawc.info()
We can see above the data structure of the multi-extension fits files. Each file contains 19 extensions which encapsulates all of the Stokes parameters in a single package.
Stokes $I$---the zeroth extension in the fits file---represents the total intensity of the image.
Let us go ahead and plot this extension:
import matplotlib.pyplot as plt
%matplotlib notebook
# ^jupyter magic for inline plots
from aplpy import FITSFigure
# set colormap for all plots
cmap = 'rainbow'
stokes_i = hawc['STOKES I'] # or hawc[0]. Note the extension is from the hawc.info() table above
axs = FITSFigure(stokes_i) # load HDU into aplpy figure
axs.show_colorscale(cmap=cmap) # display the data with WCS projection and chosen colormap
# FORMATTING
axs.set_tick_labels_font(size='small')
axs.set_axis_labels_font(size='small')
# Add colorbar
axs.add_colorbar()
axs.colorbar.set_axis_label_text('Flux (Jy/pix)')
Similarly, we can plot the Stokes Q and Stokes U images:
stokes_q = hawc['STOKES Q']
stokes_u = hawc['STOKES U']
axq = FITSFigure(stokes_q, subplot=(1,2,1)) # generate FITSFigure as subplot to have two axes together
axq.show_colorscale(cmap=cmap) # show Q
axu = FITSFigure(stokes_u, subplot=(1,2,2),
figure=plt.gcf())
axu.show_colorscale(cmap=cmap) # show U
# FORMATTING
axq.set_title('Stokes Q')
axu.set_title('Stokes U')
axu.axis_labels.set_yposition('right')
axu.tick_labels.set_yposition('right')
axq.set_tick_labels_font(size='small')
axq.set_axis_labels_font(size='small')
axu.set_tick_labels_font(size='small')
axu.set_axis_labels_font(size='small')
stokes_q = hawc['STOKES Q']
error_q = hawc['ERROR Q']
axq = FITSFigure(stokes_q, subplot=(1,2,1)) # generate FITSFigure as subplot to have two axes together
axq.show_colorscale(cmap=cmap) # show Q
axe = FITSFigure(error_q, subplot=(1,2,2), figure=plt.gcf())
axe.show_colorscale(cmap=cmap) # show error
# FORMATTING
axq.set_title('Stokes Q')
axe.set_title('Error Q')
axq.axis_labels.hide() # hide axis/tick labels
axe.axis_labels.hide()
axq.tick_labels.hide()
axe.tick_labels.hide()
Level 4 HAWC+ additionally provides extensions with the polarization percentage ($p$), angle ($\theta$), and their associated errors ($\sigma$).
Percent polarization ($p$) and error ($\sigma_p$) are calculated as:
\begin{align} p & = 100\sqrt{\left(\frac{Q}{I}\right)^2+\left(\frac{U}{I}\right)^2} \\ \sigma_p & = \frac{100}{I}\sqrt{\frac{1}{(Q^2+U^2)}\left[(Q\,\sigma_Q)^2+(U\,\sigma_U)^2+2QU\,\sigma_{QU}\right]+\left[\left(\frac{Q}{I}\right)^2+\left(\frac{U}{I}\right)^2\right]\sigma_I^2-2\frac{Q}{I}\sigma_{QI}-2\frac{U}{I}\sigma_{UI}} \end{align}
Note that $p$ here represents the percent polarization as opposed to the more typical convention for $p$ as the fractional polarization.
Maps of these data are found in extensions 7 (PERCENT POL) and 9 (ERROR PERCENT POL).
Polarized intensity, $I_p$, can then be calculated as $I_p = \frac{I\times p}{100}$, which is included in extension 13 (POL FLUX).
Also included is the debiased polarization percentage ($p^\prime$) calculated as:
$p^\prime=\sqrt{p^2-\sigma_p^2}$, found in extension 8 (DEBIASED PERCENT POL).
We similarly define the debiased polarized intensity as $I_{p^\prime} = \frac{I\times p^\prime}{100}$, which is included in extension 15 (DEBIASED POL FLUX).
These values are also included in table form in extension 17 (POL DATA).
stokes_ip = hawc['DEBIASED POL FLUX']
axi = FITSFigure(stokes_i, subplot=(1,2,1))
axi.show_colorscale(cmap=cmap) # show I
axp = FITSFigure(stokes_ip, subplot=(1,2,2), figure=plt.gcf())
axp.show_colorscale(cmap=cmap) # show Ip
# FORMATTING
axi.set_title(r'$I$')
axp.set_title(r'$I_{p^\prime}$')
axp.axis_labels.set_yposition('right')
axp.tick_labels.set_yposition('right')
axi.set_tick_labels_font(size='small')
axi.set_axis_labels_font(size='small')
axp.set_tick_labels_font(size='small')
axp.set_axis_labels_font(size='small')
From the $Q$ and $U$ maps, the polarization angle $\theta$ is calculated in the standard way:
$\theta = \frac{90}{\pi}\,\mathrm{tan}^{-1}\left(\frac{U}{Q}\right)$
with associated error:
$\sigma_\theta = \frac{90}{\pi\left(Q^2+U^2\right)}\sqrt{\left(Q\sigma_Q\right)^2+\left(U\sigma_U\right)^2-2QU\sigma_{QU}}$
The angle map is stored in extension 10 (POL ANGLE), with its error in extension 12 (ERROR POL ANGLE).
As part of the HAWC+ reduction pipeline, $\theta$ is corrected for the vertical position angle of the instrument on the sky, the angle of the HWP plate, as well as an offset angle that is calibrated to each filter configuration. $\theta=0^\circ$ corresponds to the North-South direction, $\theta=90^\circ$ corresponds to the East-West direction, and positive values follow counterclockwise rotation.
We also provide the PA of polarization rotated by $90^\circ$, $\theta_{90}$, in extension 11 (ROTATED POL ANGLE). This PA of polarization needs to be used with caution. If the measured polarization is dominated by magnetically-aligned dust grains, then the PA of polarization, $\theta$, can be rotated by $90^\circ$ to visualize the magnetic field morphology. For more details, see Hildebrand et al. 2000; Andersson et al. 2015.
We can now use the $p^\prime$ and $\theta_{90}$ maps to plot the polarization vectors. First, however, let us make a quality cut. Rather than defining a $\sigma$ cut on the polarization vectors themselves, it is more useful to define a signal-to-noise cut on total intensity, $I$, the measured quantity.
Starting with the propagated error on the polarization fraction:
\begin{equation*} \sigma_p = \frac{100}{I}\sqrt{\frac{1}{(Q^2+U^2)}\left[(Q\,\sigma_Q)^2+(U\,\sigma_U)^2+2QU\,\sigma_{QU}\right]+\left[\left(\frac{Q}{I}\right)^2+\left(\frac{U}{I}\right)^2\right]\sigma_I^2-2\frac{Q}{I}\sigma_{QI}-2\frac{U}{I}\sigma_{UI}} \end{equation*}
Let's assume the errors in $Q$, $U$, and $I$ are comparable so that there are no covariant (cross) terms in the error expansion.
Therefore, \begin{equation*} \sigma_Q = \sigma_U = \sigma_{Q,U} \\ \sigma_{QI} = \sigma_{QU} = \sigma_{UI} = 0 \end{equation*}
$\require{cancel}$ \begin{align} \sigma_p & = \frac{100}{I}\sqrt{\frac{1}{(\cancel{Q^2+U^2})}\left[\sigma_{Q,U}^2\left(\cancel{Q^2+U^2}\right)\right]+ \sigma_I^2\left(\frac{Q^2+U^2}{I^2}\right)} \\ \sigma_p & = \frac{100}{I}\sqrt{\sigma_{Q,U}^2+\sigma_I^2\left(\frac{Q^2+U^2}{I^2}\right)}= \frac{1}{I}\sqrt{\sigma_{Q,U}^2+\sigma_I^2\,p^2} \end{align}
If we assume that $p$ is relatively small (e.g. the source is not highly polarized), and that the errors in $I$ are small, then the second term ($\sigma_I^2\,p^2$) is negligible.
\begin{equation*} \sigma_p = \frac{\sigma_{Q,U}}{I} \end{equation*}
By design, the HAWC+ optics split the incident radiation into two orthogonal linear polarization states that are measured with two independent detector arrays. The total intensity, Stokes $I$, is recovered by linearly adding both polarization states. If the data is taken at four equally-spaced HWP angles, and assuming 100% efficiency of the instrument, then the uncertainty in $I$ is related to the uncertanties in $Q$ and $U$: \begin{equation*} \sigma_Q\sim\sigma_U\sim\sqrt{2}\,\sigma_I \end{equation*}
This simplifies our error on $p$ to: \begin{align} \sigma_p &\sim \sqrt{2}\frac{\sigma_I}{I} \\ \sigma_p &\sim \frac{\sqrt{2}}{\left(S/N\right)_I} \end{align}
If we desire an error in $p$ of $\sim0.5\%$, what is the required signal-to-noise in $I$?
\begin{align} \left(\mathrm{S/N}\right)_I & \sim \sqrt{2}\left(\frac{1}{\sigma_p}\right) \sim \sqrt{2}\frac{1}{0.5\%} \\ & \sim \frac{\sqrt{2}}{0.005} \sim 283 \end{align}
So, therefore if we desire an accuracy of $\sigma_p\sim0.5\%$, we require a S/N in total intensity $I$ of $\sim283$.
This S/N cut in $I$ is very conservative. In the final HAWC+ data product, the last extension, FINAL POL DATA, contains a table of values similar to POL DATA, with somewhat less restrictive quality cuts applied. This extension includes vectors satisfying the following three criteria:
Since we include maps with the full data set, you are free to decide on any quality cuts that satisfy your scientific needs.
In this next panel, we include a single quality cut on S/N > 100, by performing the following steps:
from astropy.io import fits
import numpy as np
from aplpy import FITSFigure
def make_polmap(filename, title=None, figure=None, subplot=(1,1,1)):
hawc = fits.open(filename)
p = hawc['DEBIASED PERCENT POL'] # %
theta = hawc['ROTATED POL ANGLE'] # deg
stokes_i = hawc['STOKES I'] # I
error_i = hawc['ERROR I'] # error I
# 1. plot Stokes I
# convert from Jy/pix to Jy/sq. arcsec
pxscale = stokes_i.header['CDELT2']*3600 # map scale in arcsec/pix
stokes_i.data /= pxscale**2
error_i.data /= pxscale**2
fig = FITSFigure(stokes_i, figure=figure, subplot=subplot)
# 2. perform S/N cuts on I/sigma_I, and p/sigma_p
i_err_lim = 100
mask = np.where(stokes_i.data/error_i.data < i_err_lim)
# 3. mask out low S/N vectors by setting masked indices to NaN
p.data[mask] = np.nan
# 4. plot vectors
scalevec = 0.4 # 1pix = scalevec * 1% pol # scale vectors to make it easier to see
fig.show_vectors(p, theta, scale=scalevec, step=2) # step size = display every 'step' vectors
# step size of 2 is effectively Nyquist sampling
# --close to the beam size
# 5. plot contours
ncontours = 30
fig.show_contour(stokes_i, cmap=cmap, levels=ncontours,
filled=True, smooth=1, kernel='box')
fig.show_contour(stokes_i, colors='gray', levels=ncontours,
smooth=1, kernel='box', linewidths=0.3)
# Show image
fig.show_colorscale(cmap=cmap)
# If title, set it
if title:
fig.set_title(title)
# Add colorbar
fig.add_colorbar()
fig.colorbar.set_axis_label_text('Flux (Jy/arcsec$^2$)')
# Add beam indicator
fig.add_beam(facecolor='red', edgecolor='black',
linewidth=2, pad=1, corner='bottom left')
fig.add_label(0.02, 0.02, 'Beam FWHM',
horizontalalignment='left', weight='bold',
relative=True, size='small')
# Add vector scale
# polarization vectors are displayed such that 'scalevec' * 1% pol is 1 pix long
# must translate pixel size to angular degrees since the 'add_scalebar' function assumes a physical scale
vectscale = scalevec * pxscale/3600
fig.add_scalebar(5 * vectscale, "p = 5%",corner='top right',frame=True)
# FORMATTING
fig.set_tick_labels_font(size='small')
fig.set_axis_labels_font(size='small')
return stokes_i, p, mask, fig
stokes_i, p, mask, fig = make_polmap(afile, title='A')
We can also plot the polarization fraction $p$ to better visualize the structure of 30 Doradus. We plot the same contours from total intensity $I$ in the background.
fig = FITSFigure(p)
# Show image
fig.show_colorscale(cmap=cmap)
# Plot contours
ncontours = 30
fig.show_contour(stokes_i, colors='gray', levels=ncontours,
smooth=1, kernel='box', linewidths=0.3)
# Add colorbar
fig.add_colorbar()
fig.colorbar.set_axis_label_text('$p^\prime$ (%)')
Finally, using the function defined above, we plot all four HAWC+ observations of 30 Doradus.
files = [afile,cfile,dfile,efile]
titles = ['A','C','D','E']
for file, title in zip(files,titles):
make_polmap(file,title)